EasyTensor

The official python client for EasyTensor.

Installation

Pretty straightforward.

pip install easytensor

Usage

Once you have a model exported to your local storage, you can upload it to easytensor in one line of code.

TensorFlow

Exporting and uploading a model

import easytensor
import os
export_path = os.path.join(os.getcwd(), "my_model")
print("export_path: {}".format(export_path))

# Export the model
tf.keras.models.save_model(
    model,
    export_path,
    overwrite=True,
    include_optimizer=True,
    save_format=None,
    signatures=None,
    options=None
)

# Upload it to easytensor.
model_id, access_token = easytensor.tensorflow.upload_model("My first model", export_path)
print("model ID:", model_id)
print("access token:", access_token)

Running prediction on the cloud

from pprint import pprint
import requests
response = requests.post(
    "https://app.easytensor.com/query/",
    json={
        "instances": [
            image_to_predict.numpy().tolist()
        ]
    },
    headers={"accessToken": access_token}
)
print("Response from server:")
pprint(response.json())

Examples

The library comes with a few example Jupyter notebooks that walk you through a few possible workflows. They are helpful if you are starting out with ML or remote model prediction.

Requirements

• Tensorflow 2. TF2 currently (early 2021) requires a python version 3.5-3.8. You will have to install a compatible version of python.
• virtualenv
• jupyter notebook

For Mac

brew install python@3.8

For Ubuntu

sudo apt install python3.8 python3.8-dev

To run the examples, create a python virtual env, and install jupyter notebook.

# install virtualenv
pip3 install virtualenv

# create a virtualenv with python3.8 in ~/virtualenv-3.8
virtualenv --python=$(which python3.8) ~/virtualenv-3.8

# activate the virtual env
source ~/virtualenv-3.8/bin/activate

# install jupyter notebook and necessary widgets
pip install notebook ipywidgets

# run jupyter notebook
jupyter notebook

Questions and Help

If you have any querstions about how EasyTensor works or want help with serving your ML model, please contact me directly at kamal@easytensor.com. I'm here to help!